A smart agricultural data processing method and a smart agricultural digital platform

Abstract

The application relates to the technical field of intelligent agricultural digital platforms, and discloses an intelligent agricultural data processing method and an intelligent agricultural digital platform, the method comprising the following steps: pre-processing target field image data, extracting a first parameter data set representing macroscopic growth trends of seedlings, extracting weak high-frequency texture information of early diseases and insect pests in a noise residual image of the image data as a second parameter data set representing local pathological conditions, and quantifying a pathological risk parameter through the second parameter data set; combining agricultural operation record data, performing seedling growth analysis, and outputting planting growth analysis results; and determining an agricultural operation strategy adjustment result according to the planting growth analysis results. The application is beneficial to solving the technical problem that it is difficult to give field data management and prospective prediction based on existing monitoring data in the existing agricultural technology.

Description

Technical Field

[0001] This invention relates to the field of smart agriculture digital platform technology, and in particular to a smart agriculture data processing method and a smart agriculture digital platform. Background Technology

[0002] Agriculture, as a fundamental industry, is undergoing a critical period of transformation from traditional extensive farming to modern intensive farming. However, most agricultural production management at present still relies heavily on the personal experience of growers, lacking scientific and quantifiable decision-making basis. This extensive management model has led to many insurmountable technical challenges.

[0003] First, the production end lacks precise sensing and intelligent control capabilities. In key operational stages such as soil fertility management, irrigation, and pest and disease control, producers often rely solely on past experience for judgment, making it difficult to grasp the physiological state and environmental parameters of crops in real time and dynamically. Especially for early-stage pests and diseases that are easily concealed, or subtle changes in the soil microenvironment, manual inspections suffer from significant delays and a high rate of misjudgment. This not only leads to excessive or insufficient application of water and fertilizer, resulting in resource waste and increased production costs, but also causes serious agricultural non-point source pollution (such as fertilizer runoff and soil compaction), causing large fluctuations in crop yields and inconsistent quality (such as sugar content, size, and taste), severely restricting the improvement of agricultural efficiency.

[0004] Secondly, while video surveillance technology has been introduced in some modern agricultural scenarios, there are significant gaps in the in-depth mining and utilization of image data. Current monitoring systems are primarily limited to real-time previewing and playback of video footage, with the release of data value entirely dependent on manual verification. However, faced with massive amounts of surveillance video streams, manual review is not only extremely inefficient but also prone to omissions due to visual fatigue. More importantly, real-time previewing and playback focus more on the macroscopic growth status of crops, making it difficult to extract localized characteristics of crop growth for agricultural operations. In other words, relying solely on manual review cannot accurately capture subtle localized characteristics of crops, rendering this video-based monitoring method ineffective in supporting early crop health management.

[0005] Furthermore, agricultural data is fragmented and suffers from severe spatiotemporal heterogeneity. The crop growth cycle is influenced by multiple factors, including weather and growing environment, resulting in large volumes of complex monitoring data. In existing agricultural data processing methods, crop growth image data (area data), IoT sensor data (point data), and meteorological data are often collected and stored independently. Current technologies typically rely on manual experience to qualitatively analyze the correlations between these three types of data, lacking automated quantitative analysis methods. Therefore, existing agricultural monitoring platforms struggle to effectively correlate discrete sensor values ​​with continuous crop growth images within a unified dimension, failing to accurately uncover the specific coupled impacts of environmental changes on crop growth. Consequently, they are unable to provide accurate field data management and forward-looking predictions based on existing monitoring data.

[0006] In summary, existing agricultural technologies lack the ability to deeply mine image data and deeply integrate and analyze agricultural production data, making it difficult to provide scientific field data management and forward-looking predictions based on existing monitoring data. Summary of the Invention

[0007] The main objective of this invention is to provide a smart agriculture data processing method and a smart agriculture digital platform, which aims to solve the technical problem in existing agricultural technologies that make it difficult to provide field data management and forward-looking predictions based on existing monitoring data.

[0008] To achieve the above objectives, the present invention provides a smart agriculture data processing method, comprising the following steps: Acquire multi-source monitoring data of the target field, including image data, IoT sensor data, agricultural operation record data and base meteorological data; The image data is preprocessed to extract the first parameter dataset representing the macroscopic growth status of seedlings. The image data is then subjected to edge preservation and denoising to generate a low-frequency structure image. The difference between the image data and the low-frequency structure image is calculated to obtain a noise residual image containing high-frequency details and random noise. The weak high-frequency texture information of early pests and diseases is extracted from the noise residual image as a second parameter dataset representing the local pathological condition, so as to quantify the pathological risk parameter through the second parameter dataset. The first parameter dataset, the second parameter dataset, IoT sensor data and base meteorological data are spatiotemporally aligned and fused, and combined with agricultural operation record data, seedling growth analysis is performed to output planting growth analysis results. Among them, the planting growth analysis results include seedling quality monitoring and evaluation, seedling farm planting statistical analysis and agricultural operation instructions. The results of the agricultural operation strategy adjustment were determined based on the planting growth analysis.

[0009] Optionally, the step of preprocessing the image data to extract a first parameter dataset characterizing the macroscopic growth status of the seedlings includes: The target field is divided into several sub-regions based on the geographic coordinate system. Image data of each sub-region is acquired. The image data of selected sample plots in each sub-region are processed to determine the plant height, leaf area index and biomass data of the seedlings. Based on biomass data and preset growth evaluation criteria, the distribution of seedling quality grades within the quadrat area is determined, and the number of seedlings of each quality grade in the sub-region is estimated based on the distribution of seedling quality grades within the quadrat area. Based on the area proportion of each sub-region, determine the estimated quantity of seedlings of different quality grades within the target field; A first parameter dataset is established based on the geographical coordinates of each sub-region, the estimated number of seedlings of different quality grades, plant height, leaf area index, and biomass data.

[0010] Optionally, the step of extracting weak high-frequency texture information of early pests and diseases from the noisy residual image as a second parameter dataset characterizing the local pathological condition includes: Guided filtering is used to perform edge-preserving denoising on image data. The filtering radius and gray value range variance are set to preserve the leaf edge structure and filter out random noise, generating a low-frequency structure image. The difference between the image data and the low-frequency structure image is determined to obtain the noise residual image, which includes weak high-frequency texture information caused by early pests and diseases. Wavelet packet decomposition is performed on the noisy residual image to extract high-frequency subband coefficients; The pathological risk parameters, lesion density coefficients, and early pest and disease stress levels are determined based on high-frequency subband coefficients to establish a second parameter dataset.

[0011] Optionally, the step of spatiotemporally aligning and fusing the first parameter dataset, the second parameter dataset, IoT sensor data, and base meteorological data, and combining them with agricultural operation record data to perform seedling growth analysis and output planting growth analysis results includes: Using a preset time granularity, the high-frequency collected IoT sensor data and base meteorological data are averaged and aggregated using a sliding window to achieve time alignment; Based on the geographic coordinate system, discrete IoT sensor location data are converted into a continuous spatial surface layer using Kriging interpolation to achieve spatial alignment. Based on the comparison results of the pathological risk parameters of each sub-region in the second parameter dataset with the preset threshold, the lesion density coefficient of each sub-region and the early pest and disease stress level, the spatial spread gradient of pests and diseases is determined. Based on the spatial spread gradient, and combined with the number of seedlings of each quality level in the sub-region in the first parameter dataset, the damage probability of each sub-region is calculated. Based on the damage probability of each sub-region and combined with agricultural operation record data, seedling growth analysis is performed and planting growth analysis results are output.

[0012] Optionally, the step of analyzing seedling growth and outputting planting growth analysis results based on the damage probability of each sub-region and in conjunction with agricultural operation record data includes: Based on the geographical distribution of each sub-region and the corresponding damage probability, a damage probability driving matrix is ​​constructed based on the actual damage probability. The source location of early pests and diseases in space and the damage probability gradient spreading to the surrounding areas are extracted based on the damage probability driving matrix. Obtain damage probability change data after implementing agricultural operation strategies for each sub-region, construct a damage probability change matrix, and extract the damage probability decay rate of each sub-region under the action of agricultural operation strategies based on the damage probability change matrix. The damage probability driving matrix and the damage probability change matrix are mapped and matched to calculate the gradient blocking coefficient of each sub-region under the corresponding agricultural operation strategy. Based on the gradient blocking coefficient, the effectiveness of agricultural operation strategies in inhibiting the spatial spread of pests and diseases is evaluated, and agricultural operation adjustment instructions are generated, which include strategies to strengthen operations at the source location or adjust the direction of spread gradient.

[0013] Optionally, the step of determining the results of agricultural operation strategy adjustments based on the planting growth analysis results includes: Characteristic attribution analysis was performed on the planting growth analysis results. Combined with agricultural operation record data, the multidimensional offset between the current seedling's actual growth trajectory and the preset standard growth baseline was measured. The response sensitivity of IoT sensor data and base meteorological data as environmental parameters to the multidimensional offset was analyzed to identify the associated factors that restrict growth. Based on the correlation factors and their corresponding response sensitivities, the corresponding agricultural operation control strategies are matched through a pre-set decision mapping model. Based on the agricultural operation control strategy, the target control parameters are determined, and operation control instructions containing specific execution instructions are generated; The operation control instructions are pushed to the corresponding execution terminal, and the status information fed back by the execution terminal is received.

[0014] Optionally, the step of processing the image data of selected quadrat areas within each sub-region to determine the seedling height, leaf area index, and biomass data includes: Image segmentation is performed on the image data of the sample plot area to generate a seedling mask. Edge detection is performed on the seedling mask to extract the outline of the main stem of the seedling. Based on the main stem outline, the distance from the top of the seedling to the base is measured by pixel equivalent conversion to determine the plant height; The total number of effective pixels in the seedling mask is counted, and the total number of effective pixels is mapped to the actual physical area to determine the leaf area index; Based on the circumcircle convex hull of the seedling mask, the ratio of the mask area to the circumcircle convex hull is calculated to determine the canopy coverage, and biomass data is determined based on the canopy coverage.

[0015] Optionally, the step of using guided filtering to perform edge-preserving denoising on the image data, setting the filtering radius and grayscale variance to preserve the leaf edge structure and filter out random noise to generate a low-frequency structure image includes: Set the filtering radius and regularization parameter of the guided filter, where the regularization parameter corresponds to the control factor of the variance of the grayscale range; Using image data as both the guide and input images, calculate the local mean and local variance of the image data. Based on local mean and local variance, a filtering weight module is constructed to preserve gradient information in image edge regions and smooth pixel values ​​in flat regions. The image data is processed by the filtering weighting module to output a low-frequency structure image.

[0016] Optionally, the step of determining pathological risk parameters, lesion density coefficients, and early pest and disease stress levels based on high-frequency subband coefficients includes: Threshold segmentation is performed on the high-frequency subband coefficients, and connected components exceeding the preset intensity threshold are extracted and marked as suspected lesion regions; The lesion density coefficient is determined based on the proportion of the total area of ​​suspected lesions to the total area of ​​the seedling cover. Calculate the mean energy of the high-frequency subband coefficients within the suspected lesion region to determine pathological risk parameters; Based on the lesion density coefficient and pathological risk parameters, the early pest and disease stress level is output.

[0017] To achieve the above objectives, the present invention also proposes a smart agriculture digital platform, which applies the aforementioned smart agriculture data processing method; the system includes: The data acquisition layer is used to acquire multi-source monitoring data of the target field, including image data, IoT sensor data, agricultural operation record data and base meteorological data. The data processing layer is used to preprocess the image data, extract the first parameter dataset representing the macroscopic growth status of seedlings, and perform edge preservation and noise reduction processing on the image data to generate a low-frequency structure image. The difference between the image data and the low-frequency structure image is calculated to obtain a noise residual image containing high-frequency details and random noise. The weak high-frequency texture information of early pests and diseases is extracted from the noise residual image as a second parameter dataset representing the local pathological condition, so as to quantify the pathological risk parameter through the second parameter dataset. The data analysis layer is used to perform spatiotemporal alignment and fusion of the first parameter dataset, the second parameter dataset, IoT sensor data and base meteorological data, and combine them with agricultural operation record data to perform seedling growth analysis and output planting growth analysis results. Among them, the planting growth analysis results include seedling quality monitoring and evaluation, seedling farm planting statistical analysis and agricultural operation instructions. The data application layer is used to determine the results of agricultural operation strategy adjustments based on the results of planting growth analysis.

[0018] The technical solution of this invention helps to solve the technical problem in existing agricultural technologies that make it difficult to provide field data management and forward-looking predictions based on existing monitoring data. A detailed analysis follows: This invention preprocesses image data to extract a first parameter dataset representing the macroscopic growth status of seedlings, enabling dynamic real-time perception of the overall growth status of seedlings. This replaces subjective judgment by the naked eye and eliminates reliance on growers' personal experience. Simultaneously, it utilizes edge-preserving denoising to generate low-frequency structural images and obtains noise residual images through image difference. By leveraging image noise signals that need to be removed in existing image denoising processes, it extracts weak, high-frequency texture information of early-stage pests and diseases—indistinguishable to the naked eye and uncaptured by conventional monitoring—from the noise residual images. This constructs a second parameter dataset quantifying pathological risk, overcoming the bottleneck of manual inspection or review of monitoring videos, which can only identify late-stage visible pests and diseases and suffers from delayed misjudgments. It also breaks through the limitations of traditional monitoring systems that cannot perform automatic analysis, transforming simple image monitoring data into quantifiable seedling growth and pathological risk data. This provides precise data support for subsequent agricultural operations such as water and fertilizer regulation and pest and disease prevention, contributing to stable crop yield and quality and improving agricultural production efficiency.

[0019] Furthermore, the technical solution of this invention integrates multi-source heterogeneous data, performing unified spatiotemporal alignment and fusion processing on a first parameter dataset containing seedling growth status, a second parameter dataset representing early pathological risks, field IoT sensor data, and base meteorological data. This step combines seedling status data, field environment data, and meteorological data to achieve unified integration and in-depth mining of multi-dimensional agricultural data, transforming scattered field monitoring data into a complete seedling growth data system.

[0020] Based on the fusion and in-depth mining of multi-source data, the system outputs planting growth analysis results through seedling growth analysis. Specifically, this includes seedling quality monitoring and evaluation, seedling farm planting statistical analysis, and implementable agricultural operation instructions. This enables the direct transformation of monitoring data into field management plans. At the same time, based on the comprehensive data analysis results, the system dynamically determines the adjustment results of agricultural operation strategies. It can provide forward-looking and targeted planting control plans according to the real-time growth status of seedlings, pathological risks, and environmental changes. This achieves precise control and risk prediction of key links in agricultural production, solves the core technical problems of existing technologies such as lack of quantitative decision-making basis, lack of forward-looking prediction, and extensive management, and realizes the upgrade of agricultural production from extensive management based on manual experience to data-driven intelligent management. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the smart agriculture data processing method in the first embodiment of the present invention. Figure 2 This is a simulation diagram of the target field monitoring in this invention; Figure 3 This is the seedling quality grade statistics interface in this invention; Figure 4 This is the seedling quality monitoring and evaluation interface in this invention; Figure 5 This is the dynamic interface for agricultural operations in this invention; Figure 6 This is the monthly task record statistics interface in this invention; Figure 7 This is the dynamic interface for irrigation today in this invention; Figure 8 An interface for generating dynamic prompts for each sub-region of the target field in this invention; Figure 9 This is the meteorological monitoring interface in this invention; Figure 10 This is the soil moisture monitoring interface in this invention; Figure 11 This is the seedling growth monitoring interface in this invention.

[0022] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0023] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0024] In the following description, the use of suffixes such as "unit," "component," or "element" to denote elements is solely for the purpose of illustrative purposes and has no specific meaning in itself. Therefore, "unit," "component," or "element" may be used interchangeably.

[0025] Please see Figures 1 to 11 The first embodiment of the present invention provides a smart agriculture data processing method, comprising the following steps: Step S10: Obtain multi-source monitoring data of the target field, including image data, IoT sensor data, agricultural operation record data and base meteorological data; Step S20: Preprocess the image data, extract the first parameter dataset representing the macroscopic growth status of the seedlings, and perform edge preservation and noise reduction processing on the image data to generate a low-frequency structure image. Calculate the difference between the image data and the low-frequency structure image to obtain a noise residual image containing high-frequency details and random noise. Extract the weak high-frequency texture information of early pests and diseases from the noise residual image as a second parameter dataset representing the local pathological condition, so as to quantify the pathological risk parameter through the second parameter dataset. Step S30: The first parameter dataset, the second parameter dataset, the IoT sensor data and the base meteorological data are spatiotemporally aligned and fused, and combined with the agricultural operation record data, the seedling growth analysis is performed and the planting growth analysis results are output. The planting growth analysis results include seedling quality monitoring and evaluation, seedling farm planting statistical analysis and agricultural operation instructions. Step S40: Determine the results of agricultural operation strategy adjustments based on the planting growth analysis results.

[0026] The technical solution of this invention helps to solve the technical problem in existing agricultural technologies that make it difficult to provide field data management and forward-looking predictions based on existing monitoring data. A detailed analysis follows: This invention preprocesses image data to extract a first parameter dataset representing the macroscopic growth status of seedlings, enabling dynamic real-time perception of the overall growth status of seedlings. This replaces subjective judgment by the naked eye and eliminates reliance on growers' personal experience. Simultaneously, it utilizes edge-preserving denoising to generate low-frequency structural images and obtains noise residual images through image difference. By leveraging image noise signals that need to be removed in existing image denoising processes, it extracts weak, high-frequency texture information of early-stage pests and diseases—indistinguishable to the naked eye and uncaptured by conventional monitoring—from the noise residual images. This constructs a second parameter dataset quantifying pathological risk, overcoming the bottleneck of manual inspection or review of monitoring videos, which can only identify late-stage visible pests and diseases and suffers from delayed misjudgments. It also breaks through the limitations of traditional monitoring systems that cannot perform automatic analysis, transforming simple image monitoring data into quantifiable seedling growth and pathological risk data. This provides precise data support for subsequent agricultural operations such as water and fertilizer regulation and pest and disease prevention, contributing to stable crop yield and quality and improving agricultural production efficiency.

[0027] Furthermore, the technical solution of this invention integrates multi-source heterogeneous data, performing unified spatiotemporal alignment and fusion processing on a first parameter dataset containing seedling growth status, a second parameter dataset representing early pathological risks, field IoT sensor data, and base meteorological data. This step combines seedling status data, field environment data, and meteorological data to achieve unified integration and in-depth mining of multi-dimensional agricultural data, transforming scattered field monitoring data into a complete seedling growth data system.

[0028] Based on the fusion and in-depth mining of multi-source data, the system outputs planting growth analysis results through seedling growth analysis. Specifically, this includes seedling quality monitoring and evaluation, seedling farm planting statistical analysis, and implementable agricultural operation instructions. This enables the direct transformation of monitoring data into field management plans. At the same time, based on the comprehensive data analysis results, the system dynamically determines the adjustment results of agricultural operation strategies. It can provide forward-looking and targeted planting control plans according to the real-time growth status of seedlings, pathological risks, and environmental changes. This achieves precise control and risk prediction of key links in agricultural production, solves the core technical problems of existing technologies such as lack of quantitative decision-making basis, lack of forward-looking prediction, and extensive management, and realizes the upgrade of agricultural production from extensive management based on manual experience to data-driven intelligent management.

[0029] Specifically, in the multi-source monitoring data of this invention, image data is used to monitor each sub-region of the target field and provide image information related to the geographic coordinate labels of the sub-regions; IoT sensor data is used to collect environmental physical parameters of each sub-region of the target field; and agricultural operation record data is used to obtain records of human intervention in each sub-region of the target field. This refers to the digital record of management actions performed by agricultural producers during the planting cycle, such as fertilizer type, fertilizer amount, irrigation duration, pesticide formula, pruning date, etc. Agricultural operation records such as fertilization, pruning, and transplanting in each sub-region can be entered through mobile or fixed terminals.

[0030] The system can collect RGB images of seedlings using high-definition cameras deployed in various sub-regions of the target field. It can also use camera data uploaded by agricultural producers for each sub-region as image data. In addition, drones can be used to acquire image data for each sub-region. The system collects physical environmental parameters of each sub-region through an Internet of Things sensor network, such as soil moisture, pH value, EC value (conductivity), and air temperature and humidity. All multi-source monitoring data are uniformly timestamped and labeled with the geographic coordinates of the sub-regions before being uploaded to the cloud server.

[0031] Furthermore, the first parameter dataset is a set of indicators used to describe the overall growth scale, morphological structure, and biomass data accumulation of seedlings, thereby quantifying the health status and growth progress of seedlings and realizing macro-monitoring of large-scale planting areas.

[0032] Meanwhile, the second parameter dataset is used to characterize the local pathological condition of seedlings. For seedling diseases (especially insect pests), their early development only brings subtle changes to the seedlings. Traditional image processing methods struggle to extract disease signals from these subtle changes. Furthermore, existing image processing techniques typically preprocess the acquired images, such as denoising, to remove signal jumps in isolated pixels. However, these subtle pathological signals of the seedlings themselves are also removed during denoising. Therefore, this invention preprocesses the image data. Based on the denoised image data obtained through preprocessing, a first parameter dataset representing the macroscopic growth status of the seedlings is obtained. The preprocessed noise signal is then used to sequentially generate a low-frequency result image, a noise residual image, and weak high-frequency texture information of early-stage diseases and pests, thus obtaining an early second parameter dataset characterizing the local pathological condition.

[0033] Among them, pathological conditions refer to the condition of diseases or pests.

[0034] Furthermore, once the second parameter dataset of the target field is obtained, an alarm is triggered on the monitoring system interface.

[0035] In step S30, the first parameter dataset, the second parameter dataset, the IoT sensor data, and the base meteorological data are aggregated by sliding window averaging in units of a set duration (e.g., days); all four types of data are associated with the geographic coordinates of each sub-region of the target field.

[0036] Based on a geographic coordinate system, discrete sensor point data is converted into a continuous areal layer using Kriging interpolation. The first parameter dataset, the second parameter dataset, aligned IoT sensor data, agricultural operation records (such as fertilizer types and irrigation amounts for the past 7 days), and base meteorological data are concatenated into a high-dimensional feature vector according to time series. This high-dimensional feature vector is then used to infer seedling quality monitoring and evaluation, seedling farm planting statistical analysis, and agricultural operation instructions for various seedlings in the target field. Specifically, seedling quality monitoring and evaluation reflects the pest and disease situation of the seedlings.

[0037] Furthermore, step S10 may also include: Insect traps are deployed in each sub-area of ​​the target field to form insect monitoring points and to acquire trapping data. The image data includes monitoring images inside the insect traps and automatically acquires images of insects inside the traps at preset time intervals (e.g., every 12 hours). Step S20 also includes: The system receives images of insects, uses target detection algorithms to identify and count the insects in the images, distinguishes the types of pests and counts their numbers, and obtains pest data. Based on the pest data of each sub-region, the pest density of the sub-region is calculated. Combined with the geographical location and wind direction data of each sub-region, the pest spread trend and high-risk areas of the entire target field are inferred. The pest density data and pest spread trend data of each sub-region are used as a subset of the second parameter dataset.

[0038] According to the first embodiment of the smart agriculture data processing method of the present invention, and the second embodiment of the smart agriculture data processing method of the present invention, the step S20 of preprocessing the image data and extracting the first parameter dataset characterizing the macroscopic growth status of seedlings includes: Step S21: Divide the target field into several sub-regions based on the geographic coordinate system, acquire image data of each sub-region, process the image data of the selected sample plots in each sub-region, and determine the seedling height, leaf area index and biomass data. Step S22: Based on biomass data and preset growth evaluation standards, determine the distribution of seedling quality grades within the quadrat area, and estimate the number of seedlings of each quality grade in the sub-area based on the distribution of seedling quality grades within the quadrat area. Step S23: Based on the area proportion of each sub-region, determine the estimated quantity of seedlings of different quality grades in the target field; Step S24: Based on the geographical coordinates of each sub-region, the estimated number of seedlings of different quality grades, plant height, leaf area index, and biomass data, establish the first parameter dataset.

[0039] Among them, biomass data is a growth indicator of seedlings, used to determine the quality grade of seedlings, and seedling quantity is a population quantity indicator of seedlings.

[0040] Specifically, in each sub-region, monitoring quadrat areas are selected, images are collected, and target detection is used to identify seedling masks; Wherein, the total number of seedlings in the quadrat area equals the number of mask connected components; At the same time, each type of seedling was classified into first-level, second-level, third-level and ungraded seedlings according to plant height, leaf area index and biomass data, so as to obtain the proportion of each level in the quadrat area.

[0041] Based on the area of ​​the quadrat region, the total area of ​​the subregion, and the proportion of seedlings of each grade within the quadrat region, the proportion of seedlings of each grade within the total area of ​​the subregion can be estimated.

[0042] In this embodiment, a quality evaluation matrix is ​​constructed based on plant height, leaf area index, and biomass data within the quadrat area to determine the quality grade S of various seedlings within the quadrat area. Thus, the seedlings in each quadrat area are divided into four levels: Grade 1 seedlings: Plant height, leaf area index, and biomass data all reach 90% of the standard growth baseline. Second-level seedlings: Plant height, leaf area index, and biomass data all reach 70% of the standard growth baseline and are less than 90% of the standard growth baseline. Grade III seedlings: Plant height, leaf area index, and biomass data all reach 50% of the standard growth baseline and are less than 70% of the standard growth baseline. Out-of-grade seedlings: Plant height, leaf area index, and biomass data are all less than 50% of the standard growth baseline. The proportion of seedlings of each quality grade in each quadrat area was statistically analyzed to determine the distribution of seedling quality grades within the quadrat area. Based on the distribution of seedling quality grades within the quadrat area and the area proportion of the quadrat area in the corresponding sub-region, the estimated quantity of seedlings of each quality grade in the sub-region was calculated.

[0043] In a second embodiment of the smart agriculture data processing method of the present invention, and in a third embodiment of the smart agriculture data processing method of the present invention, the step S20 of extracting weak high-frequency texture information of early pests and diseases from the noisy residual image as a second parameter dataset characterizing the local pathological condition includes: Step S25: Guided filtering is used to perform edge-preserving denoising on the image data. The filtering radius and gray value range variance are set to preserve the blade edge structure and filter out random noise, generating a low-frequency structure image. Step S26: Determine the difference between the image data and the low-frequency structure image to obtain a noise residual image, wherein the noise residual image includes weak high-frequency texture information caused by early pests and diseases; Step S27: Perform wavelet packet decomposition on the noisy residual image and extract the high-frequency subband coefficients; Step S28: Determine pathological risk parameters, lesion density coefficients, and early pest and disease stress levels based on high-frequency subband coefficients to establish a second parameter dataset.

[0044] In this invention, edge-preserving denoising refers to smoothing flat areas of an image (i.e., denoising) while preserving edge and texture details. Guided filtering is preferred to effectively avoid gradient inversion artifacts. The low-frequency structural image refers to the image that, after edge-preserving denoising, retains only the large-scale outline of objects while removing minute details and random noise. The noise residual image is the difference between the original image and the low-frequency structural image in the image data. In this invention, subtle texture changes caused by early pests and diseases (such as small scabs on leaves, chlorotic spots, and relatively obvious insect feeding marks) often manifest as weak high-frequency signals. These signals are often smoothed out in conventional denoising. However, this invention, in order to extract disease information, specifically preserves them in the residual image to identify weak high-frequency texture information—that is, minute structural changes existing in the high-frequency band of the image that are difficult to detect with the naked eye. In agricultural scenarios, this usually corresponds to microscopic pathological features.

[0045] Specifically, the system receives field image data. Due to the complex field environment, images often contain wind blur, uneven lighting, and sensor noise. First, a guided filtering algorithm is used to process the images. A filtering radius (r = 5~10 pixels) and grayscale variance are set to generate a low-frequency structural image. The purpose of this step is to remove background noise and highlight the main structure of the leaves. Then, the difference between the original image data and the low-frequency structural image is calculated to obtain a noise residual image. At this point, a large area of ​​the leaf background is canceled out, leaving only the faint high-frequency texture that is considered noise by conventional algorithms but actually contains pathological information.

[0046] For example, the tiny, meandering tunnels that leaf miners create inside leaves appear as clear, high-contrast lines in the residual image; and the edges of small, brown lesions are enhanced in the residual image for early anthracnose.

[0047] When determining pathological risk parameters based on high-frequency subband coefficients, it is necessary to calculate the average energy of the high-frequency subband coefficients. Healthy leaves have uniform texture and low energy, while diseased leaves have disordered texture and significantly increased energy. In addition, the lesion density coefficient is calculated, and the high-frequency coefficients are thresholded. Connected regions exceeding the preset intensity are extracted as suspected lesions, and the proportion of the total area of ​​suspected lesions to the area of ​​the seedling cover is statistically analyzed.

[0048] Therefore, by combining the average energy and the lesion density coefficient, the early level of pest and disease stress is determined by outputting the level through a preset mapping table (e.g., energy > 0.8 and density > 5% correspond to severe stress).

[0049] The pathological risk parameters, lesion density coefficients, and early pest and disease stress levels of each sub-region are used to construct a second parameter dataset. Furthermore, the pest density data and pest spread trend data of each sub-region can also be included as subsets of the second parameter dataset.

[0050] In the third embodiment of the smart agriculture data processing method of the present invention, and in the fourth embodiment of the smart agriculture data processing method of the present invention, step S30 includes: Step S31: Using a preset time granularity, perform sliding window averaging aggregation on the high-frequency collected IoT sensor data and base meteorological data to complete time alignment; Step S32: Based on the geographic coordinate system, the discrete IoT sensor location data is converted into a continuous spatial surface layer using Kriging interpolation to complete spatial alignment. Step S33: Based on the comparison results of the pathological risk parameters of each sub-region in the second parameter dataset with the preset threshold, the lesion density coefficient of each sub-region and the early pest and disease stress level, determine the spatial spread gradient of pests and diseases. Step S34: Based on the spatial spread gradient and combined with the number of seedlings of each quality level in the sub-region of the first parameter dataset, calculate the damage probability of each sub-region. Step S35: Based on the damage probability of each sub-region and combined with agricultural operation record data, perform seedling growth analysis and output the planting growth analysis results.

[0051] The propagation gradient is determined as follows: The comprehensive disease evaluation factor for each sub-region is obtained by weighted summation of the comparison results of the pathological risk parameters of each sub-region with the preset threshold, the lesion density coefficient of each sub-region and the early pest and disease stress level.

[0052] By connecting sub-regions in ascending order of comprehensive disease evaluation factors, the spatial spread direction of pests and diseases is obtained. Based on the comprehensive disease evaluation factors of each sub-region connected in the spread direction, the spread gradient is determined.

[0053] Specifically, the spatial spread gradient reflects the growth trend of comprehensive disease evaluation factors. Combined with the number of seedlings of each quality grade in each sub-region, the probability of damage is calculated. It is easy to understand that the greater the spread pressure and the more weak seedlings in the sub-region, the higher the probability of damage.

[0054] Right now, ; in, This represents the probability of damage. For spatial spread gradient; The seedling resistance coefficient is determined based on the proportion of first-grade seedlings to the total number of seedlings in the sub-region. The environmental stress coefficient is determined based on IoT sensor data and base meteorological data.

[0055] For example, if the current ambient humidity is greater than 80%, the temperature is within the range suitable for pathogen reproduction (e.g., 25℃-30℃), and the weather data indicates rainy weather, then the environmental stress coefficient is 1.2 (high sensitivity environment); conversely, if the current ambient humidity is less than 50%, the temperature is within the range unsuitable for pathogen reproduction (e.g., 10℃-15℃), and the weather data indicates sunny and dry weather, then the value is 0.8 (low sensitivity environment).

[0056] Based on the above analysis of damage probability and gradient blocking coefficient, the system outputs specific planting growth analysis results, which include: (1) Seedling quality monitoring and evaluation: Based on the first parameter dataset (plant height, leaf area index, and biomass data) and the second parameter dataset (pathological risk parameters), seedling quality thermal data is generated to quantify the distribution ratio of first-grade, second-grade, third-grade, and substandard seedlings in each sub-region. Combined with the damage probability, high-risk decline areas and robust growth areas are marked.

[0057] (2) Statistical analysis of seedling cultivation: Based on spatiotemporally aligned multi-source data, the system outputs growth rate curves and cumulative biomass change charts for specific periods. Statistical analysis results include: the average pass rate of the current batch of seedlings (ratio of Grade 1 to Grade 2 seedlings), the expected date of reaching the standard, and the expected yield reduction rate due to pests and diseases (estimated based on damage probability).

[0058] (3) Agricultural operation instructions: Based on the gradient blocking coefficient and the damage probability driving matrix, differentiated agricultural operation instructions are generated, including: targeted pesticide application instructions: for the source location of pests and diseases extracted from the damage probability driving matrix, high-dose, precise spraying plant protection coordinates and pesticide ratio parameters are generated; environmental control instructions: for highly sensitive environmental areas with high humidity or high temperature, start-stop control instructions for ventilation, shading, or irrigation and drainage equipment are generated to reduce the environmental stress coefficient; nutrient intervention instructions: for sub-regions with biomass data below the standard baseline and low resistance coefficient of strong seedlings, irrigation valve opening instructions for additional water and fertilizer and fertilizer application suggestions are generated.

[0059] Seedling quality monitoring and evaluation reports, seedling farm planting statistics and agricultural operation instructions are packaged together into planting growth analysis results, which are then pushed to user terminals or automated execution equipment through the data application layer to achieve closed-loop management.

[0060] In the fourth embodiment of the smart agriculture data processing method of the present invention, and in the fifth embodiment of the smart agriculture data processing method of the present invention, step S35 includes: Step S351: Based on the geographical distribution of each sub-region and the corresponding damage probability, construct a damage probability driving matrix driven by the actual damage probability, and extract the spatial source location of early pests and diseases and the damage probability gradient spreading to the surrounding areas based on the damage probability driving matrix. Step S352: Obtain damage probability change data after implementing agricultural operation strategies for each sub-region, construct damage probability change matrix, and extract the damage probability decay rate of each sub-region under the action of agricultural operation strategies based on the damage probability change matrix. Step S353: Map and match the damage probability driving matrix and the damage probability change matrix to calculate the gradient blocking coefficient of each sub-region under the corresponding agricultural operation strategy. Step S354: Based on the gradient blocking coefficient, evaluate the inhibitory effect of agricultural operation strategies on the spatial spread of pests and diseases, and generate agricultural operation adjustment instructions that include strategies to strengthen operations at the source location or adjust the direction of spread gradient.

[0061] Specifically, when constructing the damage probability driving matrix, the target field is divided into an N×M grid matrix, with each grid being a sub-region. The geographic coordinates of each sub-region are mapped to the corresponding grid nodes. The damage probability of each sub-region is used as the attribute value of the corresponding grid node, and Kriging interpolation is used to fill the grid nodes without data to generate a continuous damage probability driving matrix.

[0062] Determining the source location and damage probability gradient specifically includes: calculating the gradient magnitude and gradient direction of each grid node in the damage probability driving matrix; marking grid nodes with gradient magnitudes exceeding a preset threshold and gradient directions diverging as the source locations of early pests and diseases; generating damage probability contour lines along the gradient direction, and using the rate of change of damage probability between adjacent contour lines as the damage probability gradient.

[0063] Furthermore, the gradient blocking coefficient is calculated as follows: ; Let be the gradient blocking coefficient corresponding to the i-th row and j-th column grid, used to characterize the disease suppression efficiency under the initial spread gradient of the i-th row and j-th column grid; where... Let $\frac{i}{j}$ be the damage probability decay rate corresponding to the grid in the $i$-th row and $j$-th column. Let be the initial damage probability gradient value corresponding to the grid in the i-th row and j-th column. , .

[0064] pass It represents the potential energy for the spread of pests and diseases. This represents the inhibitory force generated by agricultural operations (such as spraying pesticides). It is calculated... This quantifies the inhibition efficiency under unit diffusion potential energy. If the initial diffusion is rapid (large denominator) but decays rapidly after agricultural operations (large numerator), it indicates that the operational strategy is precise and effective. This provides a quantitative basis for agricultural decision-making.

[0065] In the first embodiment of the smart agriculture data processing method of the present invention, and in the sixth embodiment of the smart agriculture data processing method of the present invention, step S40 includes: Step S41: Perform feature attribution analysis on the planting growth analysis results. Combine agricultural operation record data to determine the multidimensional offset between the current seedling's actual growth trajectory and the preset standard growth baseline. Analyze the IoT sensor data and base meteorological data as environmental parameters to determine the response sensitivity of the multidimensional offset, so as to identify the associated factors that restrict growth. Step S42: Based on the correlation factors and their corresponding response sensitivities, the corresponding agricultural operation control strategy is matched through a preset decision mapping model. Step S43: Based on the agricultural operation control strategy, determine the target control parameters and generate operation control instructions containing specific execution instructions; Step S44: Push the job control command to the corresponding execution terminal and receive the status information fed back by the execution terminal.

[0066] The system calls a preset standard growth baseline, which is an ideal growth curve model constructed based on the growth data of high-quality seedlings of the same variety from the same historical period. The system overlaps and compares the actual growth trajectory of the current seedling (i.e., the curve of changes in growth parameters over time as monitored in real time) with the standard growth baseline, and calculates the difference between the two in multiple dimensions (such as plant height, leaf area, and biomass data) to obtain the multidimensional offset.

[0067] Subsequently, the system analyzes IoT sensor data (such as soil moisture, pH value, EC value (electrical conductivity), air temperature and humidity, light intensity, and CO2 concentration) and base meteorological data (such as weather, rainfall, and accumulated temperature), and inputs this data as environmental parameters into the response sensitivity analysis module. This module uses a gradient-based attribution algorithm (such as DeepLift or integral gradient method) to calculate the contribution of each environmental parameter to the multidimensional offset.

[0068] For example, if the calculation shows that the sensitivity of soil moisture content to the offset of plant height growth lag exceeds the set range (i.e., the contribution exceeds the preset sensitivity threshold), the system determines that the current growth restriction is not caused by pests or diseases or insufficient fertilizer, but by water stress.

[0069] Based on this, the system identifies soil moisture content as a contributing factor to growth restriction.

[0070] After determining the correlation factors and their corresponding response sensitivities, the system performs strategy matching through a pre-set decision mapping model.

[0071] This decision mapping model stores the mapping relationship between environmental factors and agricultural operation strategies.

[0072] When the associated factor is insufficient soil moisture content and the response sensitivity indicates that the seedlings are in a water-sensitive period, the decision mapping model will match an incremental irrigation agricultural operation control strategy. If the correlation factor is insufficient accumulated temperature due to low nighttime temperatures, the model will match the strategy of covering with an insulation film or turning on the heating equipment.

[0073] Based on the determined agricultural operation control strategy, the system transforms it into specific target control parameters and generates machine-executable operation control instructions.

[0074] For example, if the strategy is incremental irrigation, the target control parameters include: irrigation area coordinates (water-scarce sub-regions determined based on geographic coordinate systems), irrigation duration (e.g., 30 minutes), and water valve opening degree (e.g., 80%).

[0075] The system encapsulates these parameters into standard control protocol instructions (such as MQTT or Modbus protocol instructions) to form job control instructions.

[0076] In the third embodiment of the smart agriculture data processing method of the present invention, and in the seventh embodiment of the smart agriculture data processing method of the present invention, the step S21 of processing the image data of the selected sample plot area in each sub-region to determine the seedling height, leaf area index, and biomass data includes: Step S211: Perform image segmentation on the image data of the sample plot area to generate a seedling mask, perform edge detection on the seedling mask, and extract the main stem outline of the seedling. Step S212: Based on the main stem outline, the distance from the top of the seedling to the base is measured by pixel equivalent conversion to determine the plant height; Step S213: Count the total number of effective pixels in the seedling mask and map the total number of effective pixels to the actual physical area to determine the leaf area index; Step S214: Based on the circumcircle convex hull of the seedling mask, calculate the ratio of the mask area to the circumcircle convex hull to determine the canopy coverage, and determine the biomass data based on the canopy coverage.

[0077] First, the acquired quadrat area image data is preprocessed, including histogram equalization to enhance contrast and to separate the seedling targets from the complex field background (such as soil, weeds, and mulch).

[0078] Specifically, the system calculates the super-green index (ExG = 2G - R - B, where ExG is the super-green index, R is the grayscale value of the Red channel of each pixel in the image, G is the grayscale value of the Green channel of each pixel in the image, and B is the grayscale value of the Blue channel of each pixel in the image), sets an adaptive threshold to binarize the image, and generates a binary mask image. In the binary mask image, foreground pixels (value 1) represent seedling areas, and background pixels (value 0) represent non-seedling areas, thus generating an accurate seedling mask.

[0079] Based on the generated seedling mask, the system uses edge detection to extract the edge contour of the seedling mask. Considering the vertical growth characteristic of seedlings, the system uses connected component analysis to select the connected region with the largest area as the candidate region for the main stem, and extracts its main stem contour line.

[0080] To determine plant height, the system establishes an image coordinate system and identifies the extreme points of the main stem outline in the vertical direction (Y-axis).

[0081] Specifically, the pixel height of the main stem outline is determined by calculating the ordinates of the highest and lowest points of the main stem outline, and the pixel height is converted into the actual plant height based on the pixel equivalent (i.e., the actual physical length represented by each pixel) pre-calibrated by the shooting device.

[0082] The total number of all foreground pixels (effective pixels) in the seedling mask is counted, and the actual physical projection area of ​​the seedling leaves is calculated based on the pixel equivalents mentioned above.

[0083] The actual physical projection area of ​​the seedling leaf is the product of the square of the number of individual pixels and the total number of foreground pixels.

[0084] The leaf area index is the ratio of the actual physical projected area of ​​the seedling leaves to the actual area occupied by the sample plot.

[0085] To estimate biomass data without loss, this embodiment introduces canopy coverage as an intermediate variable. The system calculates the circumscribed convex hull of the seedling mask. The circumscribed convex hull is the smallest convex polygon region that contains all pixels of the seedling mask.

[0086] The ratio of the actual pixel area of ​​the seedling mask to the area of ​​the circumscribed convex hull region is calculated as the canopy coverage, reflecting the fullness and spread of the seedling leaves.

[0087] A regression model of canopy coverage-biomass data is established in advance (e.g., a power function model obtained by fitting a large number of sample measured data). Substituting the calculated canopy coverage into the regression model, the current biomass data can be retrieved and output.

[0088] in, For biomass data; Canopy coverage; The proportionality coefficient, obtained by fitting a large amount of measured data, reflects the basic growth status and density characteristics of seedlings under specific growth conditions. b The allometric growth index, derived through data fitting, represents the nonlinear growth rate at which canopy coverage affects biomass data.

[0089] In the third embodiment of the smart agriculture data processing method of the present invention, and in the eighth embodiment of the smart agriculture data processing method of the present invention, step S25 includes: Step S251: Set the filtering radius and regularization parameter of the guided filter, wherein the regularization parameter corresponds to the control factor of the variance of the gray value range; Step S252: Using the image data as the guide image and input image, calculate the local mean and local variance of the image data; Step S253: Based on local mean and local variance, a filtering weight module is constructed to preserve gradient information in the image edge region and smooth pixel values ​​in flat regions. Step S254: The image data is processed by the filtering weight module to output a low-frequency structure image.

[0090] Specifically, the system first initializes the global parameters required for guided filtering, including the filter radius and regularization parameter. The filter radius determines the size of the neighborhood window used for local statistics (e.g., a 5×5 pixel window centered on the current pixel). The regularization parameter, as a control factor for the variance of the grayscale range, sets the sensitivity threshold for distinguishing between flat and edge regions in the image. A smaller regularization parameter results in more sensitive edge preservation; a larger regularization parameter results in a stronger smoothing effect on the image.

[0091] The system uses the acquired seedling image data as both the guide image and the input image, traversing every pixel in the image. For each pixel, the system determines a neighborhood window based on its corresponding filter radius, statistically analyzes the grayscale distribution of pixels within the neighborhood window, and calculates the local mean and local variance within the neighborhood window. The local mean reflects the average brightness level of the neighborhood window, while the local variance reflects the degree of grayscale variation within the neighborhood window.

[0092] Based on the calculated local mean and local variance, and combined with preset regularization parameters, the system constructs a filtering weight module. This module is used to generate filtering coefficients that adapt to local image features. In the image edge region: when the local variance value is large (i.e. the gray level changes drastically and is much larger than the regularization parameter), the filtering weight module will determine that there is a leaf edge or texture structure at this point, thereby generating a weight coefficient close to 1, so that the output pixel value retains the gradient information of the original pixel as much as possible and avoids edge blurring.

[0093] In flat regions: When the local variance is small (i.e., the grayscale changes are gentle and close to or less than the regularization parameter), the filtering weight module will determine that this is a flat region or contains random noise, and thus generate a smaller weight coefficient to smooth the region and filter out noise interference.

[0094] The system uses a pre-constructed filtering weight module to weight the raw image data. By calculating and updating each pixel in the image data with its corresponding filtering weight, the system outputs a low-frequency structural image. This image effectively filters out sensor noise and illumination clutter while clearly preserving the leaf outlines and main morphological structures of the seedlings, providing a high-quality data foundation for subsequent feature extraction.

[0095] In the seventh embodiment of the smart agriculture data processing method of the present invention, and in the ninth embodiment of the smart agriculture data processing method of the present invention, the step S28 of determining the pathological risk parameters, lesion density coefficient, and early pest and disease stress level based on the high-frequency subband coefficient includes: Step S281: Threshold segmentation is performed on the high-frequency subband coefficients, and connected components exceeding the preset intensity threshold are extracted and marked as suspected lesion regions; Step S282: Determine the lesion density coefficient based on the ratio of the total area of ​​suspected lesion areas to the total area of ​​the seedling cover. Step S283: Calculate the average energy of the high-frequency subband coefficients within the suspected lesion area to determine the pathological risk parameters; Step S284: Output the early pest and disease stress level based on the lesion density coefficient and pathological risk parameters.

[0096] In this embodiment, to further improve the adaptability of image processing, the system employs a dynamic adjustment strategy to set the value of the regularization parameter. The specific steps are as follows: When traversing image data, the system not only calculates the local mean, but also monitors the grayscale range within the current neighborhood window in real time and defines it as the local contrast.

[0097] The system has a preset contrast threshold. When the detected local contrast is higher than this threshold (usually meaning the area contains rich leaf texture or edge details), the system automatically reduces the value of the regularization parameter. This is done to reduce the smoothing effect, preserve high-frequency details of the image to the greatest extent, and prevent leaf edges from being blurred. Conversely, when the local contrast is low (meaning the area is mostly background or a flat region), the system appropriately increases the regularization parameter to enhance the noise reduction effect.

[0098] Through this dynamic adjustment mechanism based on local contrast, the present invention can simultaneously meet the requirements of strong denoising and strong edge preservation in the same image, significantly improving the accuracy of seedling phenotypic analysis.

[0099] Furthermore, based on the lesion density coefficient and pathological risk parameters, a mapping table is found to output the early pest and disease stress level.

[0100] According to the second embodiment of the smart agriculture data processing method of the present invention, and the tenth embodiment of the smart agriculture data processing method of the present invention, the method further includes: Step S50: Set the threshold range for biomass data and divide the soil moisture data into multiple moisture levels; Step S60: Establish a two-dimensional mapping matrix between biomass data and soil moisture data; Step S70: If the current seedling biomass data is within the set biomass data threshold range and the corresponding soil moisture data is lower than the preset water threshold, it is determined to be mild water shortage stress. Step S80: If the current seedling biomass data is lower than the lower limit of the set biomass data threshold range and the corresponding soil moisture data is lower than the preset water critical value, it is determined to be severe water shortage stress. Step S90: If the current seedling biomass data is lower than the lower limit of the set biomass data threshold range and there are recent excessive fertilization records in the agricultural operation record data, it is determined to be fertilizer stress. Step S100: Based on the determination result, generate water shortage stress information including water shortage level and duration.

[0101] To achieve the above objectives, the present invention also proposes a smart agriculture digital platform, which applies the aforementioned smart agriculture data processing method; the system includes: The data acquisition layer is used to acquire multi-source monitoring data of the target field, including image data, IoT sensor data, agricultural operation record data and base meteorological data. The data processing layer is used to preprocess the image data, extract the first parameter dataset representing the macroscopic growth status of seedlings, and perform edge preservation and noise reduction processing on the image data to generate a low-frequency structure image. The difference between the image data and the low-frequency structure image is calculated to obtain a noise residual image containing high-frequency details and random noise. The weak high-frequency texture information of early pests and diseases is extracted from the noise residual image as a second parameter dataset representing the local pathological condition, so as to quantify the pathological risk parameter through the second parameter dataset. The data analysis layer is used to perform spatiotemporal alignment and fusion of the first parameter dataset, the second parameter dataset, IoT sensor data and base meteorological data, and combine them with agricultural operation record data to perform seedling growth analysis and output planting growth analysis results. Among them, the planting growth analysis results include seedling quality monitoring and evaluation, seedling farm planting statistical analysis and agricultural operation instructions. The data application layer is used to determine the results of agricultural operation strategy adjustments based on the results of planting growth analysis.

[0102] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms, or by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a readable storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device to enter the methods described in the various embodiments of the present invention.

[0103] In the description of this specification, references to terms such as "one embodiment," "another embodiment," "other embodiments," or "first embodiment to Xth embodiment," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, method steps, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0104] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0105] The sequence numbers of the above embodiments of the present invention are for description only and do not represent the superiority or inferiority of the embodiments.

[0106] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A smart agriculture data processing method, characterized in that, Includes the following steps: Acquire multi-source monitoring data of the target field, including image data, IoT sensor data, agricultural operation record data and base meteorological data; The image data is preprocessed to extract the first parameter dataset representing the macroscopic growth status of seedlings. The image data is then subjected to edge preservation and denoising to generate a low-frequency structure image. The difference between the image data and the low-frequency structure image is calculated to obtain a noise residual image containing high-frequency details and random noise. The weak high-frequency texture information of early pests and diseases is extracted from the noise residual image as a second parameter dataset representing the local pathological condition, so as to quantify the pathological risk parameter through the second parameter dataset. The first parameter dataset, the second parameter dataset, IoT sensor data and base meteorological data are spatiotemporally aligned and fused, and combined with agricultural operation record data, seedling growth analysis is performed to output planting growth analysis results. Among them, the planting growth analysis results include seedling quality monitoring and evaluation, seedling farm planting statistical analysis and agricultural operation instructions. The results of the agricultural operation strategy adjustment were determined based on the planting growth analysis.

2. The smart agriculture data processing method according to claim 1, characterized in that, The step of preprocessing the image data and extracting the first parameter dataset representing the macroscopic growth status of the seedlings includes: The target field is divided into several sub-regions based on the geographic coordinate system. Image data of each sub-region is acquired. The image data of selected sample plots in each sub-region are processed to determine the plant height, leaf area index and biomass data of the seedlings. Based on biomass data and preset growth evaluation criteria, the distribution of seedling quality grades within the quadrat area is determined, and the number of seedlings of each quality grade in the sub-region is estimated based on the distribution of seedling quality grades within the quadrat area. Based on the area proportion of each sub-region, determine the estimated quantity of seedlings of different quality grades within the target field; A first parameter dataset is established based on the geographical coordinates of each sub-region, the estimated number of seedlings of different quality grades, plant height, leaf area index, and biomass data.

3. The smart agriculture data processing method according to claim 2, characterized in that, The step of extracting weak high-frequency texture information of early pests and diseases from the noisy residual image as a second parameter dataset to characterize the local pathological condition includes: Guided filtering is used to perform edge-preserving denoising on image data. The filtering radius and gray value range variance are set to preserve the leaf edge structure and filter out random noise, generating a low-frequency structure image. The difference between the image data and the low-frequency structure image is determined to obtain the noise residual image, which includes weak high-frequency texture information caused by early pests and diseases. Wavelet packet decomposition is performed on the noisy residual image to extract high-frequency subband coefficients; The pathological risk parameters, lesion density coefficients, and early pest and disease stress levels are determined based on high-frequency subband coefficients to establish a second parameter dataset.

4. The smart agriculture data processing method according to claim 3, characterized in that, The steps of spatiotemporally aligning and fusing the first parameter dataset, the second parameter dataset, IoT sensor data, and base meteorological data, and combining them with agricultural operation record data to perform seedling growth analysis and output planting growth analysis results include: Using a preset time granularity, the high-frequency collected IoT sensor data and base meteorological data are averaged and aggregated using a sliding window to achieve time alignment; Based on the geographic coordinate system, discrete IoT sensor location data are converted into a continuous spatial surface layer using Kriging interpolation to achieve spatial alignment. Based on the comparison results of the pathological risk parameters of each sub-region in the second parameter dataset with the preset threshold, the lesion density coefficient of each sub-region and the early pest and disease stress level, the spatial spread gradient of pests and diseases is determined. Based on the spatial spread gradient, and combined with the number of seedlings of each quality level in the sub-region in the first parameter dataset, the damage probability of each sub-region is calculated. Based on the damage probability of each sub-region and combined with agricultural operation record data, seedling growth analysis is performed and planting growth analysis results are output.

5. The smart agriculture data processing method according to claim 4, characterized in that, The steps for analyzing seedling growth and outputting planting growth analysis results based on the damage probability of each sub-region and combined with agricultural operation record data include: Based on the geographical distribution of each sub-region and the corresponding damage probability, a damage probability driving matrix is ​​constructed based on the actual damage probability. The source location of early pests and diseases in space and the damage probability gradient spreading to the surrounding areas are extracted based on the damage probability driving matrix. Obtain damage probability change data after implementing agricultural operation strategies for each sub-region, construct a damage probability change matrix, and extract the damage probability decay rate of each sub-region under the action of agricultural operation strategies based on the damage probability change matrix. The damage probability driving matrix and the damage probability change matrix are mapped and matched to calculate the gradient blocking coefficient of each sub-region under the corresponding agricultural operation strategy. Based on the gradient blocking coefficient, the effectiveness of agricultural operation strategies in inhibiting the spatial spread of pests and diseases is evaluated, and agricultural operation adjustment instructions are generated, which include strategies to strengthen operations at the source location or adjust the direction of spread gradient.

6. The smart agriculture data processing method according to claim 1, characterized in that, The steps for determining the results of agricultural operation strategy adjustments based on planting growth analysis include: Characteristic attribution analysis was performed on the planting growth analysis results. Combined with agricultural operation record data, the multidimensional offset between the current seedling's actual growth trajectory and the preset standard growth baseline was measured. The response sensitivity of IoT sensor data and base meteorological data as environmental parameters to the multidimensional offset was analyzed to identify the associated factors that restrict growth. Based on the correlation factors and their corresponding response sensitivities, the corresponding agricultural operation control strategies are matched through a pre-set decision mapping model. Based on the agricultural operation control strategy, the target control parameters are determined, and operation control instructions containing specific execution instructions are generated; The operation control instructions are pushed to the corresponding execution terminal, and the status information fed back by the execution terminal is received.

7. The smart agriculture data processing method according to claim 3, characterized in that, The steps of processing the image data of selected sample plots within each sub-region to determine the seedling height, leaf area index, and biomass data include: Image segmentation is performed on the image data of the sample plot area to generate a seedling mask. Edge detection is performed on the seedling mask to extract the outline of the main stem of the seedling. Based on the main stem outline, the distance from the top of the seedling to the base is measured by pixel equivalent conversion to determine the plant height; The total number of effective pixels in the seedling mask is counted, and the total number of effective pixels is mapped to the actual physical area to determine the leaf area index; Based on the circumcircle convex hull of the seedling mask, the ratio of the mask area to the circumcircle convex hull is calculated to determine the canopy coverage, and biomass data is determined based on the canopy coverage.

8. The smart agriculture data processing method according to claim 3, characterized in that, The step of using guided filtering to perform edge-preserving denoising on image data, setting the filtering radius and grayscale variance to preserve the leaf edge structure and filter out random noise to generate a low-frequency structure image includes: Set the filtering radius and regularization parameter of the guided filter, where the regularization parameter corresponds to the control factor of the variance of the grayscale range; Using image data as both the guide and input images, calculate the local mean and local variance of the image data. Based on local mean and local variance, a filtering weight module is constructed to preserve gradient information in image edge regions and smooth pixel values ​​in flat regions. The image data is processed by the filtering weighting module to output a low-frequency structure image.

9. The smart agriculture data processing method according to claim 7, characterized in that, The steps for determining pathological risk parameters, lesion density coefficients, and early pest and disease stress levels based on high-frequency subband coefficients include: Threshold segmentation is performed on the high-frequency subband coefficients, and connected components exceeding the preset intensity threshold are extracted and marked as suspected lesion regions; The lesion density coefficient is determined based on the proportion of the total area of ​​suspected lesions to the total area of ​​the seedling cover. Calculate the mean energy of the high-frequency subband coefficients within the suspected lesion region to determine pathological risk parameters; Based on the lesion density coefficient and pathological risk parameters, the early pest and disease stress level is output.

10. A smart agriculture digital platform, characterized in that, The system employs the smart agriculture data processing method as described in any one of claims 1 to 9; the system comprises: The data acquisition layer is used to acquire multi-source monitoring data of the target field, including image data, IoT sensor data, agricultural operation record data and base meteorological data. The data processing layer is used to preprocess the image data, extract the first parameter dataset representing the macroscopic growth status of seedlings, and perform edge preservation and noise reduction processing on the image data to generate a low-frequency structure image. The difference between the image data and the low-frequency structure image is calculated to obtain a noise residual image containing high-frequency details and random noise. The weak high-frequency texture information of early pests and diseases is extracted from the noise residual image as a second parameter dataset representing the local pathological condition, so as to quantify the pathological risk parameter through the second parameter dataset. The data analysis layer is used to perform spatiotemporal alignment and fusion of the first parameter dataset, the second parameter dataset, IoT sensor data and base meteorological data, and combine them with agricultural operation record data to perform seedling growth analysis and output planting growth analysis results. Among them, the planting growth analysis results include seedling quality monitoring and evaluation, seedling farm planting statistical analysis and agricultural operation instructions. The data application layer is used to determine the results of agricultural operation strategy adjustments based on the results of planting growth analysis.

Images